That image can't be generated due to Microsoft's code of conduct???

Rocket Robert 0 Reputation points
2026-10-29T02:40:03.7866667+00:00

I uploaded a family photo from the 1940's wanting to colorize it. It processed the uploaded photo and got to finishing the last details and it stopped and gave this message. "That image can't be generated due to Microsoft's code of conduct. Please start a new conversation. Learn More

Copilot said it couldn't detect the reason it was denied? Anyone have the same issue? Solution?

Microsoft Copilot | Microsoft 365 Copilot | Development
0 comments No comments

1 answer

Sort by: Newest
  1. AI answer

    2026-10-29T02:41:02.61+00:00

    Start a new conversation and request only archival restoration and conservative colorization; if the same photo is refused again, report the response through Copilot’s feedback control rather than repeatedly resubmitting it.

    Copilot scans uploaded images and prompts with automated classifiers before processing them. Those safeguards are probabilistic and can make mistakes, so this message does not necessarily mean that your family photograph actually violates the Code of Conduct. Copilot may also be unable to disclose which classifier triggered the refusal.^1^

    Retry with a narrowly defined prompt

    1. Start the new conversation requested by Copilot.
    2. Upload the original scan.
    3. Use a neutral, technical prompt such as:

      Restore this scanned 1940s family photograph for archival purposes. Remove dust and scratches, improve contrast slightly, and apply subtle, historically plausible colorization. Preserve the people’s original facial features, ages, clothing, body shapes, and expressions. Do not add, remove, or stylize anything.

    4. If that is blocked, separate the operations:
      • First: “Remove dust and scratches and improve the contrast of this archival photograph. Do not otherwise alter it.”
      • In a new request using the restored result: “Apply subtle, historically plausible colorization without changing any people or objects.”

    Microsoft’s prompting guidance recommends describing the desired result rather than referring to named people, brands, characters, or imitation styles. It also notes that image requests using a photorealistic setting experienced more blocks during testing.^2^

    Avoid unnecessary wording about ethnicity, attractiveness, age changes, bodies, violence, military affiliation, or making a person look “real.” Such details are not needed for colorization and may change how an automated classifier interprets the request.

    If the same image is still blocked

    That suggests the uploaded image itself—not merely the wording—may be triggering automated scanning. Copilot explicitly scans uploaded files and images for safety risks, and its safeguards can occasionally produce unexpected results.^1^

    Report the failed response:

    • On the Copilot web page or mobile app, select the flag icon below the response.
    • In the Copilot app, alternatively press and hold the response and select Report.
    • Explain that this is a legitimate 1940s family photograph submitted for archival restoration and colorization, and that processing stopped near completion.

    Microsoft specifically directs users to these feedback controls for harmful or unexpected Copilot behavior.^3^

    There is no documented user control for overriding the safety classifier or forcing generation. If a neutral retry and split workflow are both rejected, use a conventional photo editor’s manual colorization tools while the feedback is reviewed rather than attempting to evade the filter.


    References

    1. Transparency Note for Microsoft Copilot (for individuals) | Microsoft Support
    2. Inside Track - Helping our employees when Microsoft 365 Copilot says ...
    3. Privacy FAQ for Microsoft Copilot | Microsoft Support
    AI-generated content may be incorrect. Read our transparency notes for more information.

    Was this answer helpful?

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.